CN112069930A - Vibration signal processing method and device for improving GIS equipment fault diagnosis accuracy - Google Patents

Abstract

The invention discloses a vibration signal processing method and device for improving the fault diagnosis accuracy of GIS equipment. The method includes collecting vibration signals of GIS equipment under different health levels; dividing the vibration signal of GIS equipment into multiple samples with one cycle as the time length , construct a vibration signal data set; normalize all the samples in the vibration signal data set, and then perform a one-dimensional to two-dimensional image operation to obtain an imaged vibration signal, and obtain a vibration image data set; convert the vibration image data set Divide the training set and the test set according to the preset ratio, and construct the GIS equipment fault diagnosis model based on the convolutional neural network; normalize the real-time collected GIS equipment vibration signal and then perform the image operation to obtain the image vibration signal. Input the GIS equipment fault diagnosis model, get the current health level of the GIS equipment, and realize the GIS equipment fault diagnosis. The invention judges the specific running state of the GIS equipment and has high fault diagnosis accuracy.

Description

Vibration signal processing method and device for improving GIS equipment fault diagnosis accuracy

technical field

The invention belongs to the technical field of state monitoring and fault diagnosis of GIS equipment, and more particularly, relates to a vibration signal processing method and device for improving the accuracy of fault diagnosis of GIS equipment.

Background technique

Previous studies have shown that most serious accidents in the power system are caused by some equipment failures. In recent years, the number of gas insulated switchgear (GIS) equipment and equipment has gradually increased rapidly, and its reliability is related to the security of the power grid. run. It is of great significance to the power system to study how to model the faults of GIS equipment so as to diagnose the faults timely and accurately.

GIS equipment is an indispensable equipment in today's power system, and is widely used in high-voltage and ultra-high-voltage fields. The equipment is to seal circuit breakers, isolating switches, grounding switches, transformers, arresters, busbars, connectors and outgoing terminals in a metal grounded shell, and fill it with a certain pressure of sulfur hexafluoride (SF ₆ ) Insulating gas. The faults of GIS equipment can be divided into two categories: discharge faults and mechanical faults. At present, the research methods for discharge faults of GIS equipment mainly include pulse current method, ultra-high frequency method and gas decomposition method, while the related research on mechanical faults is still in its infancy, and the main method is vibration analysis method. The pulse current method is to determine the discharge amount of the equipment by measuring the voltage change in the circuit, thereby judging the operation state of the GIS equipment, but due to the electromagnetic pulse interference existing in the GIS equipment, the diagnosis accuracy of this method is not high; The signal in the GHz frequency band is detected to judge the operation state of the GIS equipment, but this method is difficult to accurately judge the partial discharge state; the gas decomposition method is to judge the operation state of the GIS equipment according to the decomposition products of SF ₆ gas, but this method is limited to It is used during maintenance and power outage, and it is impossible to monitor the running status of the equipment in real time. At the same time, the State Grid Corporation of China issued several guidance documents on the intelligentization of high-voltage equipment, among which condition monitoring and fault diagnosis are regarded as the key functions and difficulties of intelligent electrical appliances. Therefore, it is necessary to carry out research on the mechanical condition detection technology and diagnosis method for GIS equipment, so as to discover the latent mechanical defects inside the equipment in time, and ensure the safe and stable operation of the equipment.

It has been difficult to accurately diagnose the potential hidden dangers of GIS equipment through conventional electrical characteristic parameters. The vibration signal of GIS equipment is easily measurable state information, which can reflect the rich dynamic information of its health state. However, only obtaining the vibration signal from the GIS equipment is not enough to solve the problem of fault diagnosis. It is necessary to do a lot of post-processing on the vibration signal and then build a model to achieve the purpose of fault identification. There have been many models established by traditional machine learning methods based on vibration signals. However, most of these methods have certain limitations and have not made great breakthroughs in performance. Convolutional neural networks employ a deep architectural algorithmic structure that can learn multi-level data representations corresponding to different levels of abstraction. Convolutional neural networks have made great progress in 2D data scenarios such as computer vision, but there is no mature model for 1D vibrational signals.

SUMMARY OF THE INVENTION

Aiming at the defects of the prior art, the purpose of the present invention is to provide a vibration signal processing method and device for improving the fault diagnosis accuracy of GIS equipment, aiming to solve the difficulty of manually extracting signal features and the problem of low identification quality.

In order to achieve the above object, according to an aspect of the present invention, a vibration signal processing method for improving the fault diagnosis accuracy of GIS equipment is provided, which is characterized in that, comprising the following steps:

Step 1: According to the operating status of the GIS equipment, it is divided into different health levels, and the vibration signals of the GIS equipment under different health levels are collected;

Step 2: Divide the GIS equipment vibration signal into multiple samples with the electromagnetic force period inside a GIS equipment as the time length to construct a vibration signal data set;

Step 3, normalize all samples in the vibration signal data set, and then perform a one-dimensional to two-dimensional image operation on the normalized samples to obtain an imaged vibration signal, and obtain a vibration image data set;

Step 4: Divide the vibration image data set into a training set and a test set according to a preset ratio, and build a GIS equipment fault diagnosis model based on a convolutional neural network;

Step 5: Normalize the vibration signal of the GIS equipment collected in real time and perform an image operation to obtain the imaged vibration signal and input it into the fault diagnosis model of the GIS equipment to obtain the health level of the current GIS equipment, so as to realize the fault diagnosis of the GIS equipment.

Further, the normalization process in step 3 can be expressed as:

Among them, X=x ₁ , . . . , x _K is the original sample in the vibration signal data set, x _i represents the value of the sampling point, max( _xi ) represents the maximum value in the sampling point, min( _xi ) represents the sampling point The minimum value among the points, and K is the number of sampling points of the sample.

Further, the imaging operation in step 3 specifically includes:

将振动信号数据集从直角坐标系映射到极坐标系：Step 3.1, the normalized vibration signal dataset samples are expressed as

Map the vibration signal dataset from Cartesian to polar coordinates:

是极角，r_i是极径；Among them, K is the number of sampling points of the sample, i is the ith sampling point,

is the polar angle, and _ri is the polar diameter;

来表示，其数学描述如下式所示；Step 3.2, define an inner product operation, using the symbol

to represent, its mathematical description is shown in the following formula;

运算，得到类Gram矩阵G：Step 3.3, carry out the vibration signal data set in the polar coordinate system

Operation to get the Gram-like matrix G:

Step 3.4, convert the elements in the matrix G into pixel values, and arrange them according to the positions in the matrix to obtain a vibration image data set.

Further, the training of the convolutional neural network-based GIS equipment fault diagnosis model in step 4 includes:

Step 4.1, build a convolutional neural network, take the vibration image training set as input, and output the health level of GIS equipment;

Step 4.2, train the GIS equipment fault diagnosis model based on the convolutional neural network, and select the cross entropy as the loss function of the training;

Step 4.3, test the GIS equipment fault diagnosis model based on the convolutional neural network, input the vibration image test set into the trained convolutional neural network model, get the predicted health level, and then compare the predicted health level with the real health level For comparison, the prediction accuracy is calculated, which is used to evaluate the accuracy of the model.

The research object of this method is the vibration signal collected by the outer casing of the circuit breaker when the GIS equipment is running, which can realize real-time and efficient diagnosis of mechanical faults of the GIS equipment, and the accuracy of the traditional method of fault diagnosis using vibration signals is effectively improved. promote. Based on the above problems, we have studied a fault diagnosis method for GIS equipment based on imaged vibration signals, which can convert one-dimensional vibration signals into two-dimensional images without losing the original features. Make full use of many excellent models in the field of image recognition (such as convolutional neural network models). The fault diagnosis method of GIS equipment based on image vibration signal effectively solves the difficulty of manually extracting signal features and the problem of low identification quality.

According to another aspect of the present invention, a vibration signal processing device for improving the fault diagnosis accuracy of GIS equipment is provided, characterized in that it includes:

The vibration signal acquisition module is used to divide the GIS equipment into different health levels according to the operating status of the GIS equipment, and collect the vibration signals of the GIS equipment under different health levels;

The vibration signal building module is used to divide the vibration signal of the GIS equipment into multiple samples with the electromagnetic force period inside a GIS equipment as the time length to construct the vibration signal data set;

The vibration image acquisition module is used to normalize all the samples in the vibration signal data set, and then perform the one-dimensional to two-dimensional image operation on the normalized samples to obtain the imaged vibration signal and obtain the vibration image data set;

The diagnostic model building module is used to divide the vibration image data set into a training set and a test set according to a preset ratio, and construct a GIS equipment fault diagnosis model based on a convolutional neural network;

The fault diagnosis module is used to normalize the vibration signal of the GIS equipment collected in real time and then perform an image operation to obtain the imaged vibration signal and input it into the fault diagnosis model of the GIS equipment to obtain the health level of the current GIS equipment and realize the failure of the GIS equipment. diagnosis.

Further, the normalization process can be expressed as:

Among them, X=x ₁ , . . . , x _K is the original sample in the vibration signal data set, x _i represents the value of the sampling point, max( _xi ) represents the maximum value in the sampling point, min( _xi ) represents the sampling point The minimum value among the points, and K is the number of sampling points of the sample.

Further, the imaging operation specifically includes:

将振动信号数据集从直角坐标系映射到极坐标系：The normalized vibration signal dataset samples are expressed as

Map the vibration signal dataset from Cartesian to polar coordinates:

是极角，r_i是极径；Among them, K is the number of sampling points of the sample, i is the ith sampling point,

is the polar angle, and _ri is the polar diameter;

来表示，其数学描述如下式所示；Define an inner product operation, using the notation

to represent, its mathematical description is shown in the following formula;

运算，得到类Gram矩阵G：The vibration signal data set in the polar coordinate system is

Operation to get the Gram-like matrix G:

Convert the elements in the matrix G into pixel values and arrange them according to the positions in the matrix to obtain the vibration image data set.

Through the above technical solutions conceived by the present invention, compared with the prior art, the GIS equipment fault diagnosis method based on the imaged vibration signal and the convolutional neural network overcomes the limitations of the traditional fault diagnosis method in real-time and accuracy. The invention first normalizes the collected vibration signal of the circuit breaker when the GIS equipment is running, and then uses the image operation to process the vibration signal to obtain the image vibration signal. This method can better preserve The absolute time relationship in the vibration signal, and the correlation of time information is also extracted. Then, the convolutional neural network is used to identify the imaged vibration signals of the GIS equipment under different operating states, which can judge the specific operating state of the GIS equipment and have high fault diagnosis accuracy. good performance.

Description of drawings

Fig. 1 is the schematic flow chart of the vibration signal processing method that improves the fault diagnosis accuracy rate of GIS equipment provided by the present invention;

Fig. 2 is the concrete realization flow chart of the vibration signal imaging operation provided by the present invention;

FIG. 3 is a schematic structural diagram of a convolutional neural network provided by the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

A GIS equipment fault diagnosis method based on image vibration signal and convolutional neural network, as shown in Figure 1, includes the following steps:

Step 1, according to the operating status of the GIS equipment, divide its health status into four health levels: poor, attention, good, and excellent, and collect the vibration signals of the GIS equipment under different health levels;

Step 2: Divide the collected vibration signal of the GIS equipment into multiple samples with the electromagnetic force cycle inside a GIS device as the time length, wherein the sample length is 400 sampling points, and divide the same amount of samples with different health levels to construct the vibration signal. signal dataset;

Step 3, normalize all samples in the vibration signal data set, and then perform a one-dimensional to two-dimensional image operation on the normalized samples to obtain an imaged vibration signal, and obtain a vibration image data set;

Step 4: Divide the vibration image data set into training set and test set according to the ratio of 7:3, and build a GIS equipment fault diagnosis model based on convolutional neural network, wherein the training set is used to train the GIS equipment fault based on convolutional neural network. Diagnostic model, the test set is used to test the classification accuracy of the GIS equipment fault diagnosis model;

Step 5: Normalize the vibration signal of the GIS equipment collected in real time and perform an image operation to obtain the imaged vibration signal and input it into the fault diagnosis model of the GIS equipment to obtain the health level of the current GIS equipment, so as to realize the fault diagnosis of the GIS equipment.

计算公式为：Further, the preprocessing in the step 1 is normalization processing,

The calculation formula is:

Among them, X=x ₁ , . . . , x _K is the original sample in the vibration signal data set, x _i represents the value of the sampling point, max( _xi ) represents the maximum value in the sampling point, min( _xi ) represents the sampling point The minimum value among the points, and K is the number of sampling points of the sample.

Further, the imaging operation in step 3 is shown in Figure 2, which specifically includes the following sub-steps:

其中K为样本的采样点个数，将下标作为横坐标，

的值即振幅作为纵坐标。然后将振动信号数据集从直角坐标系映射到极坐标系：Step 3.1, assume that the normalized vibration signal dataset samples are expressed as

Among them, K is the number of sampling points of the sample, and the subscript is used as the abscissa,

The value of is the amplitude as the ordinate. Then map the vibration signal dataset from the Cartesian coordinate system to the polar coordinate system:

是极角，r_i是极径。Among them, K is the number of sampling points of the sample, i is the ith sampling point,

is the polar angle, and _ri is the polar diameter.

来表示，这种新型内积可以充分利用来自两个采样点的信息，其数学描述如下式所示；In step 3.2, a new type of inner product operation is defined, using the notation

to represent that this new inner product can make full use of the information from the two sampling points, and its mathematical description is shown in the following formula;

Step 3.3, perform a novel inner product operation on the new sequence encoded by polar coordinates to obtain a Gram-like matrix;

Step 3.4: Multiply the elements in the matrix G by 256 to convert them into pixel values, and arrange them according to the positions in the matrix to obtain a vibration image data set.

Further, the training of the GIS equipment fault diagnosis model based on the convolutional neural network in the step 4 includes:

Step 4.1, build a convolutional neural network, take the vibration image training set as input, and output the health level of GIS equipment. The basic hierarchical structure of the convolutional neural network is an input layer for data input, a convolution layer for feature extraction, a ReLu function excitation layer, a pooling layer for feature selection, and a fully connected layer for classifying features. , and its structure is shown in Figure 3. The convolution layer is set to 3 layers, the pooling layer is set to 4 layers, and the number of fully connected neurons is set to the number of GIS equipment health levels. Set the hyperparameters in the convolutional neural network, set the size of the convolution kernel in the convolutional layer to 3*3, the size of the convolutional kernel in the pooling layer to 2*2, and the volume of the first convolutional layer. The number of convolution kernels is set to 8, the number of convolution kernels of the second convolution layer is set to 16, and the number of convolution kernels of the third convolution layer is set to 32.

Step 4.2, train the GIS equipment fault diagnosis model based on the convolutional neural network. Cross-entropy is selected as the loss function for training, the network learning rate is set to 0.001, the network parameters are updated using the Adam optimizer, the batch size is set to 64, and a total of 200 rounds of training are used.

Step 4.3, test the GIS equipment fault diagnosis model based on convolutional neural network. Input the vibration image test set into the trained convolutional neural network model to get the predicted fitness level. The predicted fitness level is then compared with the true fitness level to calculate the prediction accuracy, which is used to evaluate the accuracy of the model.

Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, etc., All should be included within the protection scope of the present invention.

Claims (7)

1. a vibration signal processing method that improves GIS equipment fault diagnosis accuracy, is characterized in that, comprises the following steps: Step 1: According to the operating status of the GIS equipment, it is divided into different health levels, and the vibration signals of the GIS equipment under different health levels are collected; Step 2, the GIS equipment vibration signal is divided into a plurality of samples with the electromagnetic force period inside a GIS equipment as the time length, and the vibration signal data set is constructed; Step 3, normalize all samples in the vibration signal data set, and then perform a one-dimensional to two-dimensional image operation on the normalized samples to obtain an imaged vibration signal, and obtain a vibration image data set. ; Step 4, dividing the vibration image data set into a training set and a test set according to a preset ratio, and constructing a GIS equipment fault diagnosis model based on a convolutional neural network; Step 5: Normalize the vibration signal of the GIS equipment collected in real time and perform an image operation to obtain the imaged vibration signal and input it into the fault diagnosis model of the GIS equipment to obtain the health level of the current GIS equipment, so as to realize the fault diagnosis of the GIS equipment. 2. vibration signal processing method according to claim 1 is characterized in that, in step 3, normalization processing can be expressed as:

Among them, X=x ₁ , . . . , x _K is the original sample in the vibration signal data set, x _i represents the value of the sampling point, max( _xi ) represents the maximum value in the sampling point, min( _xi ) represents the sampling point The minimum value among the points, and K is the number of sampling points of the sample. 3. The vibration signal processing method according to claim 1, wherein the imaging operation in step 3 specifically comprises: Step 3.1, the normalized vibration signal dataset samples are expressed as

Map the vibration signal dataset from Cartesian to polar coordinates:

Among them, K is the number of sampling points of the sample, i is the ith sampling point,

is the polar angle, and _ri is the polar diameter; Step 3.2, define an inner product operation, using the symbol

to represent, its mathematical description is shown in the following formula;

Step 3.3, carry out the vibration signal data set in the polar coordinate system

Operation to get the Gram-like matrix G:

Step 3.4, convert the elements in the matrix G into pixel values, and arrange them according to the positions in the matrix to obtain a vibration image data set. 4. vibration signal processing method according to claim 1, is characterized in that, the training in described step 4 is based on the GIS equipment fault diagnosis model of convolutional neural network comprises: Step 4.1, build a convolutional neural network, take the vibration image training set as input, and output the health level of GIS equipment; Step 4.2, train the GIS equipment fault diagnosis model based on the convolutional neural network, and select the cross entropy as the loss function of the training; Step 4.3, test the GIS equipment fault diagnosis model based on the convolutional neural network, input the vibration image test set into the trained convolutional neural network model, get the predicted health level, and then compare the predicted health level with the real health level For comparison, the prediction accuracy is calculated, which is used to evaluate the accuracy of the model. 5. A vibration signal processing device for improving the fault diagnosis accuracy of GIS equipment, characterized in that it comprises: The vibration signal acquisition module is used to divide the GIS equipment into different health levels according to the operating status of the GIS equipment, and collect the vibration signals of the GIS equipment under different health levels; The vibration signal building module is used to divide the vibration signal of the GIS equipment into a plurality of samples with the electromagnetic force period inside a GIS equipment as the time length to construct a vibration signal data set; The vibration image acquisition module is used to normalize all the samples in the vibration signal data set, and then perform a one-dimensional to two-dimensional image operation on the normalized samples to obtain an imaged vibration signal, and obtain Vibration image dataset; A diagnostic model building module is used to divide the vibration image data set into a training set and a test set according to a preset ratio, and construct a GIS equipment fault diagnosis model based on a convolutional neural network; The fault diagnosis module is used to normalize the vibration signal of the GIS equipment collected in real time and then perform an image operation to obtain the imaged vibration signal and input it into the fault diagnosis model of the GIS equipment to obtain the health level of the current GIS equipment and realize the failure of the GIS equipment. diagnosis. 6. The vibration signal processing device according to claim 5, wherein the normalization process can be expressed as:

Among them, X=x ₁ , . . . , x _K is the original sample in the vibration signal data set, x _i represents the value of the sampling point, max( _xi ) represents the maximum value in the sampling point, min( _xi ) represents the sampling point The minimum value among the points, and K is the number of sampling points of the sample. 7. The vibration signal processing device according to claim 5, wherein the imaging operation specifically comprises: The normalized vibration signal dataset samples are expressed as

Map the vibration signal dataset from Cartesian to polar coordinates:

Among them, K is the number of sampling points of the sample, i is the ith sampling point,

is the polar angle, and _ri is the polar diameter; Define an inner product operation, using the notation

to represent, its mathematical description is shown in the following formula;

Perform the ⊕ operation on the vibration signal data set in the polar coordinate system to obtain the Gram-like matrix G:

Convert the elements in the matrix G into pixel values and arrange them according to the positions in the matrix to obtain the vibration image data set.

Images